ComfyUI

One line: ComfyUI (140k+ stars on GitHub) is a node-based AI image & video generation studio — you wire workflow nodes together (models, prompts, samplers) and queue renders; the official Windows portable runs after extraction, macOS installs from source into your home folder.

What you get

• A local, node-based workflow canvas for image and video generation — everything runs on your machine
• Official workflow templates for hundreds of models (SDXL, Flux, Wan video, …) via comfy.org/workflows
• Huge model ecosystem: checkpoints drop into models/, or point ComfyUI at folders you already have (extra_model_paths.yaml)
• Optional cloud "partner nodes" (e.g. Google's Nano Banana) inside the same canvas

Before you start (2 minutes)

• Disk: the Windows portable is a ~2 GB download and several GB extracted; models add many more GB — keep 20 GB+ free
• Windows route: Windows 10/11 x64 — the portable bundles Python 3.13 + PyTorch (CUDA 13.0), works on NVIDIA 20-series-and-up or CPU only (other GPUs: AMD/Intel use the manual install)
• macOS route: Python 3.10–3.13 (python3 --version) and a few GB for PyTorch + dependencies
• Not needed: administrator rights, Docker, a separate Python install on Windows

Install

Route A: install with AiBOX (recommended)

1. Open AiBOX → app directory → ComfyUI → Install
2. AiBOX downloads the official artifact (sha256-verified; CN traffic auto-uses the accelerated channel):

• Windows: extracts the official portable 7z into %LOCALAPPDATA%\Programs\ComfyUI_windows_portable with the built-in tar — zero admin, no Python needed
• macOS: unpacks the source archive into ~/ComfyUI, creates .venv and installs requirements.txt

1. First run (below) → the browser opens at http://127.0.0.1:8188

Route B: manual install (official channels)

• Windows portable (official README): download ComfyUI_windows_portable_nvidia.7z → extract with 7-Zip or Explorer → double-click run_nvidia_gpu.bat (run_cpu.bat under ComfyUI\advanced for CPU-only)
• macOS / any platform (manual): clone or download the source, then:

cd ComfyUI
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/python main.py        # → http://127.0.0.1:8188

• Official desktop app: comfy.org/download (installer for Windows/macOS) — an alternative if you prefer it
• comfy-cli (official): pip install comfy-cli then comfy install

Configuration

1. Models: put checkpoints in ComfyUI/models/checkpoints (portable: ComfyUI_windows_portable\ComfyUI\models\checkpoints); video/VAE/LoRA go in their own subfolders — the UI shows the expected layout
2. Reuse existing model folders: copy extra_model_paths.yaml.example → extra_model_paths.yaml and edit the paths (README)
3. Workflows: start from the built-in default workflow, or import templates from comfy.org/workflows
4. GPU note (Windows portable): ships PyTorch CUDA 13.0 — update NVIDIA drivers if it won't start; AMD/Intel/Apple follow the manual-install sections of the README

Verification: what "works" looks like

1. http://127.0.0.1:8188 opens the node canvas with the default workflow loaded
2. A model from models/checkpoints is selectable in the loader node (first use requires downloading a model — several GB)
3. Press Queue Prompt → the node graph runs → a generated image appears
4. Version check (installed by AiBOX):

cat ~/ComfyUI/comfyui_version.py            # macOS: __version__ = "x.y.z"
type "%LOCALAPPDATA%\Programs\ComfyUI_windows_portable\ComfyUI\comfyui_version.py"   # Windows

FAQ

• Nothing opens / port busy: ComfyUI defaults to port 8188; another instance or a stale process may hold it — close it or start with --port 8189
• tar cannot extract the 7z (Windows): install 7-Zip and extract manually — same result, still zero admin
• Black screen / crash on start (portable): update the NVIDIA driver (CUDA 13.0 build), or grab the cu126 portable for older cards
• Which portable do I need: nvidia (20-series+, or CPU), nvidia_cu126 (10-series and older), amd, intel — all on the releases page
• python3 too old (macOS): install Python 3.12 from python.org (per-user, no admin) and re-run the venv steps
• Huge downloads are slow (CN): set PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple before installing requirements

Uninstall

• Windows: delete %LOCALAPPDATA%\Programs\ComfyUI_windows_portable (portable — no installer entry)
• macOS: delete ~/ComfyUI (venv + source live together)
• ⚠️ Models, workflows and custom nodes live inside those folders and can be tens of GB — back them up first

More: README install section · docs.comfy.org · workflow templates